Middleware for Sensor Networks

Transcription

1 Middleware for Sensor Networks Krzysztof Piotrowski

2 Background Application Middleware Sensor Network Application Middleware Sensor Network Middleware for Sensor Networks 2

3 Middleware layer that: Hides (abstracts) the details of the underlying distributed system: System heterogeneity, Data exchanges, etc. Simplifies programming of distributed applications Provides value-added services: Naming, Transactions, etc. Middleware for Sensor Networks 3

4 Middleware and Sensor Networks Sensor Network A distributed sensing network with a large number of devices Constraints resources A great amount of data Middleware for Sensor Networks Scalable Self-organizing Energy efficient Middleware for Sensor Networks 4

5 Data handling Measurement Input from sensor, e.g., temperature Computation Result of an internal operation, e.g., average value of several last temperature measurements used to eliminate errors Event Predefined or dynamically defined exceptional situation described based on available measurements or computations, e.g., temperature (or average) higher than 200 C Query Asynchronous access to data not available locally Middleware for Sensor Networks 5

6 Data Storage schemes External Storage Local Storage Data-Centric Storage Middleware for Sensor Networks 6

7 External Storage Middleware for Sensor Networks 7

8 Local Storage Middleware for Sensor Networks 8

9 Data-Centric Storage Data items are named with keys DCS supports two operations: PUT(k, v) stores the value v according to the key k its name GET(k) retrieves a value associated with the key k Hash function Hashes a key k into geographic coordinates PUT() and GET() use the same hash function, i.e., the same key k results in the same location Middleware for Sensor Networks 9

10 Data-Centric Storage example Put( elephant, data) (11, 28) PDA (11,28)=Hash( elephant ) Middleware for Sensor Networks 10

11 Data-Centric Storage example Get( elephant ) (11, 28) PDA (11,28)=Hash( elephant ) Middleware for Sensor Networks 11

12 Data-Centric Storage example 2 elephant PDA fire Middleware for Sensor Networks 12

13 Data Storage schemes - comparison n is the number of nodes Asymptotic costs of O(n) for floods and O(n 1/2 ) for point-to-point routing ES LS DS Cost for Storage O(n 1/2 ) 0 O(n 1/2 ) Cost for Query 0 O(n) O(n 1/2 ) Cost for Response 0 O(n 1/2 ) O(n 1/2 ) Middleware for Sensor Networks 13

14 Macro-programming WSN - Kairos In Kairos, a programmer writes a single sequential program using a simple centralized memory model General-purpose programming facility for tasking an entire collection of sensors as a single entity at a high level of abstraction Such a facility is called macro-programming Can express global behavior succinctly Centralized Sensor State mapped from Sensors Sequential Program Thread of control Read/write Middleware for Sensor Networks 14

15 Kairos idea behind & example Centralized sequential programs easier to specify, code, understand and debug than hand-coded distributed versions Reuse textbook algorithms for sophisticated tasks Ignoring latency and energy considerations, a dumb but obviously trivial distributed implementation always possible, by shipping sensor nodes state to and from a central location Example: to build a shortest path tree rooted at Root, the centralized program must capture the global behavior: For each node i in the network, its parent is that neighbor whose distance to Root is shortest Middleware for Sensor Networks 15

16 Kairos architecture Centralized Program Kairos preprocessor + language compiler Annotated Localized Binary Copies of remote managed objects Program Thread of control Sensor Node Kairos runtime Cached Objects Managed Objects Link + distribute to runtime Program Thread of control Link + distribute to runtime Sensor Node Kairos runtime Cached Objects Managed Objects Link + distribute to runtime Exported to remote nodes by the Kairos runtime sync read/write Queue Manager sync read/write Queue Manager Requests Replies Requests Replies Multi-hop wireless network Middleware for Sensor Networks 16

17 Kairos features Three constructs with which to write programs node a first-class datatype, node_list (iterator on nodes) that facilitate topology independent programming, get_neighbors() to obtain current one-hop neighbors of a node, var@node to synchronously access data and program state of nodes These constructs are language-agnostic They can be implemented in the preprocessor stage of compilation Middleware for Sensor Networks 17

18 Kairos example //Every sensor node has a dist_from_root integer variable and a parent node variable void build_tree (node root=0) { node n, n'; //get the current list of all available nodes in the network node_list avail_n=get_available_nodes(); for (;;) { for (n=get_first(avail_n); n!=null; n=get_next(avail_n)) { node_list neigh_n=get_neighbors()@n; for (n'=get_first(neigh_n); n'!=null; n'=get_next(neigh_n)) { if (dist_from_root@n'<dist_from_root@n+1) { parent@n=n'; dist_from_root@n=dist_from_root@n'+1; }}}}}} (dist_from_root=inf, parent=-1) (dist_from_root=1, parent=0) n n 1 0 (dist_from_root=0, parent=0) 4 (dist_from_root=inf, parent=-1) (dist_from_root=1, parent=0) (dist_from_root=inf, parent=-1) (dist_from_root=2, parent=1) n 2 3 (dist_from_root=inf, parent=-1) (dist_from_root=2, parent=4) Middleware for Sensor Networks 18

19 Kairos routing tree performance Compared against OPP (baseline) OPP (One-Phase-Pull) proposed for TinyDiffusion because it is traffic efficient Flood interests (requests), unicast responses (data) Measure Convergence Time, Overhead, and Stretch OPP doesn t necessarily produce shortest paths, so quantify stretch Root Middleware for Sensor Networks 19

20 Kairos routing tree performance 25 Convergence Time (S) Time OPP Time Kairos Kairos < 1.3x OPP Number of nodes Middleware for Sensor Networks 20

21 Kairos routing tree performance Overhead (bytes) Overhead Kairos Overhead OPP Kairos < 2x OPP Number of Nodes Middleware for Sensor Networks 21

22 Maté: A Tiny Virtual Machine TinyOS component 7286 bytes code, 603 bytes RAM Three concurrent execution contexts Stack-based bytecode interpreter Code broken into 24 instruction capsules Self-forwarding code Rapid reprogramming Message receive and send contexts Three execution contexts Clock, Receive, Send Seven code capsules Clock, Receive, Send, Subroutines 0-3 One word heap gets/sets instructions Two-stack architecture Operand stack, return address stack Middleware for Sensor Networks 22

23 Maté Architecture Subroutines Events Clock Send Receive Maté gets/sets PC Code Operand Stack Return Stack Mate Context Middleware for Sensor Networks 23

24 Maté Instructions One byte per instruction Three classes: basic, s-type, x-type basic: data, arithmetic, communication, sensing s-type: used in send/receive contexts x-type: embedded operands basic s-type x-type 00iiiiii 01iiixxx 1ixxxxxx i = instruction x = argument Middleware for Sensor Networks 24

25 Code Snippet: cnt_to_leds gets # 0x1b # Push heap variable on stack pushc 1 # 0xc1 # Push 1 on stack add # 0x06 # Pop twice, add, push result copy # 0x0b # Copy top of stack sets # 0x1a # Pop, set heap pushc 7 # 0xc7 # Push 0x0007 onto stack and # 0x02 # Take bottom 3 bits of value putled # 0x08 # Pop, set LEDs to bit pattern halt # 0x00 # Middleware for Sensor Networks 25

26 Maté Capsules Hold up to 24 instructions Fit in a single TinyOS AM packet Installation is atomic Four types: send, receive, clock, subroutine Context-specific: send, receive, clock Called: subroutines 0-3 Version information Middleware for Sensor Networks 26

27 Maté Contexts Each context associated with a capsule Executed in response to event external: clock, receive internal: send (in response to sendr) Execution model preemptive: clock non-preemptive: send, receive Every instruction executed as TinyOS task Middleware for Sensor Networks 27

28 Maté Viral Code Every capsule has version information Maté installs newer capsules it hears on network Motes can forward their capsules (local broadcast) forw forwo Middleware for Sensor Networks 28

29 Maté Case Study: GDI Great Duck Island application Simple sense and send loop Runs every 8 seconds low duty cycle 19 Maté instructions, 8K binary code Energy tradeoff: if you run GDI application for less than 6 days, Maté saves energy Middleware for Sensor Networks 29

30 Middleware for Sensor Networks 30

31 Kairos idea behind & example Centralized sequential programs easier to specify, code, understand and debug than hand-coded distributed versions To build a shortest path tree rooted at Root, the centralized program must capture the global behavior: For each node a in the network, its parent is that neighbor whose distance to Root is shortest Middleware for Sensor Networks 31